Technical Note: An hybrid parallel implementation for EGSnrc Monte Carlo user codes.

Area of Science:

• Computational physics
• Medical physics
• High-performance computing

Background:

• The EGSnrc Monte Carlo code system is widely used for radiation transport simulations.
• Existing parallelization methods for EGSnrc often rely on batch-queueing systems (BQS), which can limit efficiency on modern high-performance computing (HPC) architectures.
• There is a need for more efficient and flexible parallelization strategies to accelerate complex simulations.

Purpose of the Study:

• To develop and present a novel parallel solution for the EGSnrc Monte Carlo code system.
• To combine Message Passing Interface (MPI) and Open Multi-Processing (OpenMP) programming models for enhanced simulation performance.
• To offer an alternative to the existing BQS-based implementation for EGSnrc.

Main Methods:

• Leveraged a prior OpenMP implementation of EGSnrc.
• Integrated MPI features to enable efficient simulation distribution across HPC systems.
• Utilized user-defined macros controlled by compilation flags to manage parallelization strategies.
• Benchmarked the hybrid solution using the DOSXYZnrc code with a clinical photon beam and water phantom.

Main Results:

• Validated the parallel solution against serial runs, confirming no alteration in dose distribution.
• Demonstrated efficient workload distribution across available computing units.
• Achieved a 10% speedup compared to an MPI-only solution and a 30% speedup over the BQS implementation.

Conclusions:

• The presented hybrid MPI/OpenMP method offers a viable and efficient approach for parallelizing EGSnrc Monte Carlo simulations on distributed computing systems.
• This solution effectively utilizes available HPC resources.
• Users gain flexibility to choose between OpenMP-only, MPI-only, or a combined parallelization scheme.